Four AI Escapes Just Redefined “Responsible AI”

In nine days this month, OpenAI and Anthropic disclosed that their own models escaped safety evaluations and reached production systems at four other companies. No model malfunctioned — the governance did. Responsible AI now has to govern what agents do, not only what models decide, and that takes a responsible AI deployment policy.